Author Affiliations
Key Laboratory of Photonics Information Technology, Ministry of Industry and Information Technology, School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, Chinashow less
Fig. 1. Structure of long-distance optical fiber pre-warning system
Fig. 2. Φ-OTDR vibration signal acquisition system
Fig. 3. Vibration event classification process
Fig. 4. Structure of the improved neural network
Fig. 5. Multi-scale decomposition tree of wavelet packet decomposition
Fig. 6. ANN network used for classification
Fig. 7. DNN network used for classification
Fig. 8. Three signal samples in Baoshan area of Shanghai. (a) NI event; (b) MG event; (c) ME event
Fig. 9. Comparison of signal samples before and after Max Abs Scaler. (a1)(a2) NI event; (b1)(b2) MG event; (c1)(c2) ME event
Fig. 10. Signal and its WPE distributions. (a)NI event; (b)MG event; (c) ME event
Fig. 11. Relationship between number of epoch and validation accuracy, training loss during training. (a) Validation accuracy; (b) training loss
Fig. 12. Comparison of occurrence probability of MD event in model recognition. (a) Before using the improved neural network; (b) after using the improved neural network
Type of event | Number oftraining sets | Collection location oftraining sets /km | Number oftest sets | Collection location oftest sets /km |
---|
NI | 5500 | 2.99 | 2000 | 2.99 | MG | 5500 | 6.21 | 2000 | 6.21 | ME | 5500 | 5.68 | 2000 | 5.68 |
|
Table 1. The first experiment data
Type of event | Method 1 | Method 2 | Method 3 |
---|
NI | 98.2 | 95.9 | 95.0 | MG | 97.5 | 95.5 | 93.1 | ME | 96.1 | 94.2 | 92.5 | Average accuracy | 97.2 | 95.2 | 93.5 |
|
Table 2. Signal recognition accuracy of the first experimentunit:%
Type of event | Number oftraining sets | Collection location oftraining sets /km | Number oftest sets | Collection location oftest sets /km |
---|
NI | 5500 | 2.99 | 2000 | 14.96 | MG | 5500 | 6.21 | 2000 | 16.86 | ME | 5500 | 5.68 | 2000 | 6.24 |
|
Table 3. The second experiment data
Type of event | Method 1 | Method 2 | Method 3 |
---|
NI | 98.0 | 95.2 | 94.0 | MG | 96.3 | 94.0 | 92.1 | ME | 95.4 | 92.7 | 90.5 | Average accuracy | 96.5 | 93.8 | 92.2 |
|
Table 4. Signal recognition accuracy of the second experimentunit:%
Number of test sets | Method 1 | Method 2 | Method 3 |
---|
2000 | 1.74 s | 5.43 s | 1.88 s |
|
Table 5. Comparison of the recognition time of the three methods